Keywords:
Artificial Intelligence, CT, Computer Applications-General, Technology assessment, Forensics, Quality assurance
DOI:
10.26044/ecr2024/C-10469
Results
A identification using CT images was achievable for approximately 72% to 87% (rank 1-10) of the identities. In all CT regions (A-F), same-individual identification achieved a score of 12.04 ± 0.86%, while different individuals scored 2.15 ± 0.40% (refer to Figure 2 and Table 1). This resulted in successful identification rates of 60 ± 8% (rank 1), 70 ± 8% (rank 5), and 74 ± 9% (rank 10). The maxillary sinus (region E) exhibited the highest success rates at 72% (rank 1), 80% (rank 5), and 87% (rank 10). Using OPGs as a reference yielded success rates of 86% (rank 1), 87% (rank 5), and 88% (rank 10).
The higher scores for different individuals in CT slices, 5-9 times higher than in OPGs, may be attributed to factors such as metal artifacts, lower image resolution, and similar objects, like the CT table in the image.
Figure 3 illustrates the relationship between the score and the time between acquisitions. Even for images taken several years apart, successful identification was achieved.
In Figure 4, an example of found matching points is depicted. While teeth also exhibit many matching points (compare with Figure 4 A), metal artifacts disrupt the clear identification (compare with Figure 4 D). However, the skull bone and cavities such as the maxillary sinus and ear canal also provide many distinctive features that can be recognized (compare with Figure 4 E-F). Even the nose and chin can exhibit distinctive features.